You are on page 1of 15

M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P.

Perera

NanoAgents: Molecular Docking Using Multi-Agent Technology

M.Harindra R. Fernando bquharindra@gmail.com


Department of Computational Mathematics
University of Moratuwa
Sri Lanka

Asoka S. Karunananda askarunananda@gmail.com


Department of Computational Mathematics
University of Moratuwa
Sri Lanka

Roshan P. Perera rpperera@gmail.com


The Institute for Combinatorial Advanced Research and Education
General Sir John Kotelawala Defense University
Sri Lanka

Abstract

Traditional computer-based simulators for manual molecular docking for rational drug discovery
have been very time consuming. In this research, a multi agent-based solution, named as
NanoAgent, has been developed to automate the drug discovery process with little human
intervention. In this solution, ligands and proteins are implemented as agents who pose the
knowledge of permitted connections with other agents to form new molecules. The system also
includes several other agents for surface determination, cavity finding and energy calculation.
These agents autonomously activate and communicate with each other to come up with a most
probable structure over the ligands and proteins, which are participating in deliberation. Domain
ontology is maintained to store the common knowledge of molecular bindings, whereas specific
rules pertaining to the behaviour of ligands and proteins are stored in their personal ontologies.
Existing, Protein Data Bank (PDB) has also been used to calculate the space required by ligand
to bond with the receptor. The drug discovery process of NanoAgent has exemplified exciting
features of multi agent technology, including communication, coordination, negotiation, butterfly
effect, self-organizing and emergent behaviour. Since agents consume fewer computing
resources, NanoAgent has recorded optimal performance during the drug discovery process.
NanoAgent has been tested for the discovery of the known drugs for the known protein targets. It
has 80% accuracy by considering the prediction of the correct actual existence of the docked
molecules using energy calculations. By comparing the time taken for the manual docking
process with the time taken for the molecular docking by NanoAgent, there has been 95%
efficiency.

Keywords: Molecular Docking, Multi-agent, Drug Discovery.

1. INTRODUCTION
Molecular docking is a well-recognized branch of research in drug discovery under the field of
structural molecular biology. This area is primarily concerned with binding a smaller molecules
(ligand) with a target macromolecule (protein). Over many decades, the molecular binding
process has been manually modeled by scientist with reference to huge databases of protein and
ligand [1]. Nevertheless, the manual molecular docking has been a very time consuming and
tedious task. In 1990s, some computer-based solutions for molecular docking have begun to
emerge [2]. However, these solutions have appeared as semi-automated approaches to
molecular docking, and expected the scientists to use many additional software tools to complete
the docking process. As such, automated and integrated software solutions for molecular docking

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 37
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

have become a research challenge. This paper present our Multi-Agent Systems based software
solution, NanoAgent, for integrated and automated molecular docking. The NanoAgent provides
automated all-in-one molecular docking system with the facility to energy minimization, active site
prediction, cavity discovery on the target, and orientation calculations. According to experimental
results, NanoAgent shows 80% accuracy and 95% efficiency.

The rest of the research paper is organized as follows: Section II critically review the domain of
molecular docking by highlighting current solutions, practices, technologies, limitations for
defining the research problem. Section III and Section IV describes the essentials of Multi-Agent
Technology showing its relevance to solve the molecular docking domain with the aid of
Ontology. From the Section V to Section VII, this paper presents our novel approach to molecular
docking with Multi-Agent Technology. Section VIII reports on evaluation of the novel solution by
explaining evaluation strategy. Section IX concludes the outcome of the research with a note on
further work.

2. STATE OF THE ART OF THE MOLECULAR DOCKING


Molecular Docking is usually accomplished among a small molecule and a target macromolecule.
There are mainly two types of molecular docking, called protein-ligand docking and protein-
protein docking. Ligand-protein docking is the predominant among the two types of molecular
docking [3]. The protein is referred to as the target and also called as the macromolecule. Smaller
molecule is called as the drug (or "ligand") is being docked. The goal of molecular docking is to
calculate the correct binding mode of a ligand with the protein [4]. Search is performed through
the high dimensional spaces efficiently, to find out the correct binding pose. The scoring functions
can be used to rank the candidate docking results [5, p. -8] . It is required to evaluate the actual
existence of the docked structures using energy values. Most of the time scientists used to do all
the above-mentioned calculations manually, for the molecular docking. They have recognized the
advantage of the computer to simulate some steps of the manual molecular docking such as
energy calculations, cavity finding and surface matching.

The field of molecular docking has appeared during the last decades, driven by the needs of
structural molecular biology and structure-based drug discovery [6] using the computer. In
computer aided drug design and structural molecular biology heavily uses the molecular docking
software tools. In manual molecular docking, scientist needs to get the assistance from the
various combinations of separate set of computer-aided tools to generate molecular data
including geometries, energies and properties. Manual molecular docking is tedious and time-
consuming process due to these reasons.

DOCK 1.0 was the first automated receptor-ligand docking program. It was purposed in 1982 by
Irwin Kuntz in The Department of Pharmacology at The University of California at San Francisco
[7]. It is used to achieve docking which is the process that two molecules fit together in 3D space.
It has the limitations in terms of docking accuracy and time consumption [8]. Molecular docking
program, AutoDock [9] was initially written in FORTRAN-77 in 1990 by David S. Goodsell [10] in
Arthur J. Olson's laboratory with the assistance of the concepts from DOCK 1.0. It was proposed
to achieve automated docking of ligands, which are the small molecules like a candidate drug to
their macromolecular targets such as proteins. AutoDock, which is the most famous commercial
molecular docking tool [11] in the current computer aided docking context. But it doesnt have
efficient search methods and accurate empirical free energy scoring functions. It also need the
assistance from other external tools such as MGLTools [12] and FastGrid [13] to perform some
calculations.

The college of Pharmacy in Rosland Franklin University of Medicine and Science in UK [14]
which is very famous for the advance chemistry related researches, is relied heavily on
computational molecular docking tools. Nevertheless, according to them it is not much easy to
use the current available molecular docking tools [15], because they have to access multiple
numbers of tools to perform various steps of the docking operations, to the experiment only one

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 38
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

single molecular docking process. Scientists from Rosland Franklin University, also mentioned
that the regularly, 3D structure data are available for the shape of a protein and a drug
independently, but not for the two together [16]. So they have to use-docking tools to visualize the
docked molecular structures but they are not efficient and accurate enough [17].

Pattern recognition, machine learning and other artificial intelligence approaches also play a
significant role for the screening and identification of candidate molecules [18]. Artificial Neural
Network, Fuzzy logic and Genetic Algorithms has been used by most of the drug design in-silico,
but it takes considerable amount of time to find the correct orientations of the molecules. And also
they need more powerful computer processors to the calculations. In the present most of the
molecular docking systems are rather semi-automated, and do not provide integrated solutions.
This paper presents our fully automated and integrated molecular docking solution, which exploits
the power of Multi Agent Systems and Ontological engineering in modern computing.

3. MULTI-AGENT TECHNOLOGY AND ONTOLOGICAL ENGINEERING


Multi-Agent System technology provides a new paradigm for development of software solutions
[19]. This technology supports for the development of software that operates in environments that
are distributed and interconnected. Agent technology uses to automating the software tools and
the agent behavior relies on the symbolic manipulation [20] of knowledge in order to perform
meaningful operations that simulate intelligent abilities. Ontologies can denote domain vocabulary
of the agents, the conceptualization that the terms in the vocabulary are intended to capture [21].
For increasing the explanation of the knowledge structure, reducing the language ambiguities and
enabling knowledge sharing in software projects are the main usages of the ontologies [22].

3.1 Agent Technology


In the 1990s agent based systems are one of the most inspiring areas of research and
development to have emerged in information communication technology [23]. Many researches
believe that agents represent the most important new paradigm for software development
subsequently object orientation [24]. An agent is a computer system that has the ability of flexible
autonomous action in dynamic and unpredictable situations. The performance of Agent is
measure based on the environment sequences. Environments are categorized along several
dimensions such as Observable, Deterministic, Episodic, Static and Discrete [25]. An agent
perceives its environment through sensors and acts on this environment using actuators.
Following FIGURE 1 shows the relationship between the Agent and the Environment.

FIGURE 1: PEAS of Agent.

3.2 Characteristics of Software Agent


An agent system is essentially a component system exhibiting several of the characteristics as
described follows. There are five characteristics work together to make agent oriented systems
more flexible and robust to change [26].
Adaptability - An agents behavior may be improved after it has been deployed
Autonomy - An agent is responsible for its own thread of control and can pursue its own
goal largely independent of messages sent from other agents

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 39
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

Collaboration - Agents communicates and works supportively with other agents to form
multi - agent systems working together on some task
Knowledgeable - An agent is capable of reasoning about its goals and knowledge
Mobility - The ability for an agent to move from one executing context to another

3.3 Emergent Behavior of Multi-Agent Systems


Emergent behavior is a common phenomenon in complex systems. In both natural and artificial
complex systems, individual components perform their actions and make decisions based on
local information. But the whole system demonstrates properties and behaviors that have strong
global features [27]. Emergent behavior is an fundamental feature in multi-agent systems and
plays a important role in the applications of agent technology.

3.4 Distributed Decision Making of MAS


Multi-agent systems (MAS) describe the relationship between several decision-making agents.
They are implemented to solve difficult problems in aggregating the strength of each single agent
[28]. Hence, multi-agent systems are naturally of a distributed decision making nature. More
precisely, MAS is also called distributed artificial intelligence [29] emphasizing the fact that
complex problems are solved by a population of different agents. Each of them having its own
skills, knowledge, and tasks.

3.5 The Role Of Ontologies


Ontology is a computational model of some portion of the domain [30]. It is a form of a semantic
network whose nodes are concepts or objects and its arcs represent relationships or associations
among the concepts [31]. It consists of properties and attributes with constraints, functions, and
rules that control the behavior of the concepts. Following FIGURE 2 shows the simple ontology
that represents relationships and the properties of entities.

FIGURE 2: Structure of Ontology.

Agent systems are plugged with the domain ontologies. It can be used as the operating rule set
[32] of the agents. It is important to exchange and sharing knowledge in multi-agent systems [33]
and ontologies are the best way to achieve it.

4. THEORIES BEHIND THE MOLECULAR DOCKING


In molecular docking the small molecule called Ligand usually interconnects with proteins binding
sites and also required to follow some rules. Binding sites are areas of protein consider being
active in forming of compounds. These are commonly called binding modes. It also forecasts the
stability of the binding and the energy of the complex.

4.1 Steps of Molecular Docking


I. Current molecular docking tools follow the common steps as bellow:[34][35]
II. Get the complex coordinates (i.e. from the PDB).
III. Clean the complex (delete all the water and the solvent molecules and all non-interacting
ions).
IV. Add the missing hydrogen/side chain atoms and minimized the complex.
V. Clean the minimized complex (delete all the water and the solvent molecules and all non-

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 40
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

interacting ions).
VI. Detached the minimized complex in protein (lock) and ligand (key).
VII. Prepare the docking appropriate files for protein and ligand (pdb files).
VIII. Evaluate the docking results

4.2 Lock and Key Theory


Around 1890, Emil Fischer suggested a model called the lock-and-key model. It explained how
biological systems function [36] when docking small molecule with the large molecule. A ligand
fits into the active site of a protein, similar as the key fits into a lock is depicted as the following
FIGURE 3.

FIGURE 3: Lock and Key Model.

4.3 Energy Minimization


Most of the time the energy minimization analyses are done for the ligand. Ligand can exist in
various conformations, which means the shapes or the structures. The conformation, which has
minimized energy, will be the most stable conformation of the ligand [37].
There are two types of energies:

Equation (1) is used to calculate Electrostatic Potential Energy [38] . It is Electrostatic Potential
Energy a pair wise total of columbic interactions as described in equilibrium.

qi qk
Eelectrostatic = (1)
pairsnonbonded Drik

qi and qk - the one point charges


rik distance between the point charges
D - Coulomb's constant

Equation (2) is to calculate Vander Waals Potential Energy. For general behavior of non-bonded
interactions is regularly modeled by the following equation.

Aik Cik
EVanderWaals = 12 6 (2)
pairsnonbonded rik rik

r - The distance between two atoms having charges qi and qk

Vander Waals potential which expresses the interaction energy using the atom-type dependent
constants A and C. Values of A and C may be determined by a variety of methods, like non-
bonding distances in crystals and gas-phase scattering measurements.

4.4 Bond Energy


Equation (3) is to calculate the Bond Energy. The energy E, is a function of the atomic locations
R, of all the atoms in the structure. These are usually expressed in term of coordinates.

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 41
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

Ebonded - sum of internal, or bonded


Enon-bonded - sum of external or non bonded

V (R) = Ebonded + Enonbonded (3)

Equation (4) is to find the bonded energy. The Ebonded term is a sum of three terms as follows:

Ebonded = Estrech + Ebend + Erotate (4)


There are three types of atom movements as shown in the following FIGURE 4.

FIGURE 4: Molecular Interactions in Methane Molecule.

4.4.1 Stretching
Equation (5) is to find the bond energy for stretching. This is the approximation of bond energy
as a function of b0 and Kb, determines the strength of the bond.

b0 - The ideal bond length


Kb - The force constant

2
Estreching = K b ( b b0 )
1,2 pairs
(5)

4.4.2 Bending
Bond energy for bending can be calculated using Equation (6).

0 - harmonic potential
K - Based on chemical type of atoms constituting the angle

K ( )
2
Estreching = 0
(6)
angles

4.4.3 Rotating
Equation (7) can be used to calculate bond energy for rotating.

- Dihedral angle between each atom axis


Coefficient of symmetry n=1,2,3

Erotation = K (1 cos ( n )) (7)


1,4 pairs

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 42
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

4.5 Lipinski's Rule of Five


Lipinski's rule of five [39] is a rule to evaluate drug likeness or determine if a chemical compound
with a certain biological activity has properties that would make it a likely orally active drug in
living beings. Christopher A. Lipinski articulated it in 1997, based on the inspections that most
orally administered drugs are relatively small molecules.

An orally active drug has no more than one disruption of the following:
I. Should not exceed 5 hydrogen bond donors
a. The total number of nitrogenhydrogen and oxygenhydrogen bonds
II. Should not exceed 10 hydrogen bond acceptors
a. All nitrogen or oxygen atoms
III. A molecular mass should less than 500 Daltons
According to this rule had chosen the appropriate ligands for the docking.

4.5 Force Field Function


The basic functional form of potential energy in molecular mechanics includes bonded terms for
interactions of atoms that are linked by covalent bonds, and non-bonded, terms that describe the
long-range electrostatic and van der Waals forces. Following equation (8) [40] is used to calculate
the total of the above-mentioned energies:

(8)

rij - The distance between protein atom i and ligand atom j


Aij and Bij - The VDW parameters
qi and qj - The atomic charges
(rij) is usually set to 4rij , reflecting the screening effect of water on electrostatic interactions

5. MAS APPROACH TO MOLECULAR DOCKING


Our new intelligent multi-agent systems technology based molecular docking solution has been
named as NanoAgents. The proposed system get the Protein Data Bank (PDB) files as the
inputs of the ligand and the proteins and output the discovered structures of the matched protein-
ligand pairs. Ligands and proteins are programmed as the agents of the Multi-Agent System and
the rule set for the binding codified as domain ontology. Each agent operated according to the
rule set. There are the other agents, who help for the docking of LigandAgents and ProteinAgent
such as EnergyCalculationAgent, ActiveSiteFindingAgent, SurfaceMatchingAgent and
CavityFindingAgent.

5.1 Inputs
Multi-Agent system for Molecular Docking has been design to accept multiple inputs coming from
different entities of the Molecular Docking process. The following are the main inputs to the
system. TABLE 1 shows the inputs from corresponding entities.

Input Entity (Agent)


Rules for binding Ontology Agent
Description of protein Protein Agent
Description of ligand Ligand Agent
Description of bonds Energy Calculation Agent
3D coordinates of the protein Active Site Finding Agent
3D coordinates of the ligand Surface Matching Agent
3D coordinates of the ligand and protein Cavity Finding Agent
TABLE 1: Inputs for the Molecular Docking System.

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 43
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

5.2 Outputs
The output of the system will be the docked molecule. These outputs are coming as the rational
drug with its PDB file structure.

5.3 Process for the Molecular Docking


System will use the molecular of database and the ontology as the inputs to generate the rational
drugs. In this processing, two major types of agents namely protein agents and ligand agents are
defined into the system. The knowledge requires to these agents to operate are stored in
common domain ontology and personal PDB files. Get the complex coordinates (i.e. from the
PDB) as the input to the system and clean the complex (delete all the water and the solvent
molecules and all non-interacting ions).

As the next step, add the missing hydrogen/side chain atoms and minimized the complex. Then
clean the minimized complex (delete all the water and the solvent molecules and all non-
interacting ions). Separate the minimized complex in macromolecule (lock) and ligand (key)
should be done. Then prepare the docking suitable files for LOCK and KEY (pdb files). Then
perform the docking and analyze the docking results. There are separate ontologies to store the
atom colors and charges for the atoms/elements. Agents get the help from these ontologies to
find the best ligand agent to bind with the protein agent.

5.4 JADE Java Agent Development Framework


JADE (Java Agent DEvelopment Framework) [41] is a software Framework completely
implemented in the Java language. It make easy the implementation of multi-agent systems
through a middleware that compatible with the FIPA [42] specifications. JADE is completely
implemented in Java language and the minimal system requirement is the version 5 of JAVA (the
run time environment or the JDK).

5.5 Jena A Semantic Web Framework for Java


Jena [43] is a Java framework for building Semantic Web applications. It provides a development
environment for RDF [44], RDFS [45] and OWL [46], SPARQL [47] and includes a rule - based
inference engine.

5.6 Jmol: An Open-Source Java Viewer for Chemical Structures


Jmol [48] is an open-source viewer for three-dimensional chemical structures. It is used as a
research tool in chemistry and bio-chemistry. It is free and open source software written in Java
programming language and so it runs on any operating system.

6. ANALYSIS AND DESIGN OF THE NANOAGENTS


NanoAgents system design comprises of different system components for different tasks such as,
find the active site of the receptor, find the shape of the cavity of the receptor, search matching
ligand to the cavity, find correct orientation of the cavity, calculate binding energies and energy
minimization and evaluate the rationality of the selected ligand or the drug.

6.1 System Integration of the NanoAgents


The main architecture behind this solution is the blackboard architectural model that has Request
Agents, Resource Agents, Message Space and Domain Ontology. AI researchers conceived the
concept of blackboard architecture [49] in the 1970's.A message space, which is the blackboard,
has all the solutions given by the agents. The common knowledge base, the "blackboard", is
iteratively updated by a diverse group of specialist knowledge sources. The overall system
architecture and the control flow are showed as following Fig 4.

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 44
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

FIGURE 5: System Architecture.

Usually it starts with a problem description and ending with a solution. Each knowledge source
revises the blackboard with a partial solution when its internal restrictions match the blackboard
state and the specialists work together to solve the problem. The blackboard model was originally
designed as a way to handle complex, ill-defined problems, where the solution is the sum of its
parts. Molecular docking can be consider as a complex problem which the solution is the sum of
different operations, such as energy calculations, active site finding, surface matching and cavity
finding.

7. DETAILED DESIGN AND IMPLEMENTATION


In this section let us discuss the software realization of our designed system. And also in here the
system flows and algorithm level realizations will be discussed. Implementation comprises of the
multi-agent systems based molecular docking system.

7.1 DOCK Algorithm To Find Matching Ligand To The Protein


DOCK is another molecular docking program currently using in the drug discovery field. Following
are the steps it follows to find the matching ligand, achieve the docking [50]. And also we have
followed their approach similarly.

I. A set of spheres is created inside the active sties of the protein


II. The spheres represent the volume which could be occupied by the ligand
III. Ligand is represented by spheres inside the ligand (see FIGURE 6)

FIGURE 6: DOCK program algorithm: Match the spheres shown in gray (distances between them are used
for scoring).

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 45
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

7.2 Methods to Find Active Sites of The Protein


7.2.1 Alpha Spheres
Alpha shape [51] of a set of weighted vertices. It is a concept from computational geometry that
has found its use in computational chemistry. The sphere that contacts some atoms on its
boundary and contains no internal atoms is called a contact sphere. An alpha sphere is a
distinctive case of a contact sphere. An alpha sphere is a sphere that contacts four atoms on its
boundary and contains no internal atoms as FIGURE 7.

FIGURE 7: An Alpha sphere in 2D that contacts three atoms of different radius.

7.2.2 Finding Cavities in The Protein


One of many properties we can study is a molecular cavity, where cavity is understood as a free
space inside a molecule. After reviewing the literature, we are usually interested only in a certain
subset of cavities. Voronoi diagram is then used to achieve the task of finding cavities for varying
radiuses [52]. In this research it is used this method to find cavities in the target molecule.

FIGURE 8: Cavities Formed by Gray Atoms.

8. EVALUATION
8.1. Evaluation Methods
For the evaluation of the tool, compared known ligand-protein pair docking time taken in the usual
manner with this software tool. Also analyzed the time difference for measuring the efficiency of
the tool. The accuracy of the tool had been measured by the energy calculations. It will help to
find the actual existence of the ligand-protein pairs in the environment found by the docking tool.

Some of the following ligand and pairs found from the reference [53] had been used to the
evaluation of the docking tool as the sample population.

Protein Ligand
1A52 OHT
1A52 EST
2AM9 B5R
1A52 GEN
1EWV KAI
4PVU BRL
4P6W DEX
TABLE 2: Known Protein-Ligand Pairs.

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 46
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

To compare the results, used that time taken for the manual molecular docking for each protein
and ligand pairs. The next step was tracking the time taken for each docking process for the
same protein-ligand pairs taken by the developed tool. To evaluate the accuracy of the software
tool, also noted that the prediction of the actual existence of each docked pair using the tools
energy calculation module.

8.2 Test Results


Following are the two criteria used to evaluate the tool as explained the above paragraph:

Criteria 01:
The time taken for the molecular docking tool and the manual molecular docking
Criteria 02:
The prediction of the actual existence of the docked molecule

Precision for Criteria 01:


Considered the time taken for the molecular docking tool and the manual molecular docking for
the evaluation of the tool.

Probability of success (time taken by the software is lesser than the actual manual molecular
docking) = 19/20 = 0.95

Probability of failure (time taken by the software is higher than the actual manual molecular
docking) = 1/20 = 0.05

Accuracy of the software = 9/20 = 0.95 = 95%

Precision for Criteria 02:


Considered the prediction accuracy of the actual existence of the docked molecules by the
software tool by using energy calculation.

Probability of success (software predicts the actual existence of the docked molecule correctly) =
16/20 = 0.8

Probability of failure (software not predicts the actual existence of the docked molecule correctly)
= 4/20 = 0.2

Accuracy of the software = 16/20 = 0.8 = 80%

Following FIGURE 8 shows the main functions of the molecular docking tool and the interfaces.
Considering the main features of the software it is an all-in-one solution for the molecular docking
comparing other popular tool AutoDock.

FIGURE 8: Shows the Cavities of The Protein.

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 47
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

Following FIGURE 9 shows the ligand and the receptor on the same window.

FIGURE 9: Ligand and Receptor.

9. CONCLUSION AND FUTURE WORK


In this research we could show multi-agent technology can be successfully applying to automated
the manual molecular docking. The time taken to dock the molecules using our tool was
compared with the popular tool called AutoDOCK. And the results were significant.

I have executed the molecular simulation 20 times for the known ligand-protein pairs and
compared it with the actual time taken for the manual molecular docking. The experiment results
derived based on two scenarios.

Time for the docking: By comparing the time taken for the manual docking process with the time
taken for the automated molecular docking, it says the software tool can obtain 95% accuracy
Accuracy: It has 80% accuracy by considering the prediction of the correct actual existence of the
docked molecules according to the energy calculations.

So it is evident that the software can manage a good balance in between the efficiency and the
accuracy.

As the future work we can improve the tool using Artificial Neural Network by training the system.
It will again enhance the efficiency of the tool by reducing the time taken for the docking process.
Currently in this research, it was focused on the ligand-protein docking only, but it can be
extended to the protein-protein docking as well. This tool can be used by the researches in
Biological Chemistry domain. Hope to get their feed back to improve the accuracy of the software
further.

10. REFERENCES
[1] D. T.-H. Chang, C.-H. Ke, J.-H. Lin, and J.-H. Chiang, AutoBind: automatic extraction of
proteinligand-binding affinity data from biological literature, Bioinformatics, vol. 28, no. 16,
pp. 21622168, Aug. 2012.

[2] M. Munetomo, K. Akama, and H. Maeda, An automated ligand evolution system using
Bayesian optimization algorithm, WSEAS Trans. Inf. Sci. Appl., vol. 6, no. 5, pp. 788797,
2009.

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 48
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

[3] S. S. Azam and S. W. Abbasi, Molecular docking studies for the identification of novel
melatoninergic inhibitors for acetylserotonin-O-methyltransferase using different docking
routines, Theor. Biol. Med. Model., vol. 10, p. 63, Oct. 2013.

[4] G. M. Morris and M. Lim-Wilby, Molecular docking, in Molecular Modeling of Proteins,


Springer, 2008, pp. 365382.

[5] M. A. D. JE, Virtual screening and lead optimisation to identify novel inhibitors for HDAC-8,
ArXiv Prepr. ArXiv12092793, 2012.

[6] V. Lounnas, T. Ritschel, J. Kelder, R. McGuire, R. P. Bywater, and N. Foloppe, CURRENT


PROGRESS IN STRUCTURE-BASED RATIONAL DRUG DESIGN MARKS A NEW
MINDSET IN DRUG DISCOVERY, Comput. Struct. Biotechnol. J., vol. 5, no. 6, pp. 114,
Feb. 2013.

[7] Molecular Docking to Ensembles of Protein Structures, Ronald M.A. Knegtel,Irwin


D.Kuntz,Department of PharmaceuticalChemistry School of Pharmacy University of
California, San Francisco, CAand, C.M.Oshiro, 1997. .

[8] Z. Zhou, A. K. Felts, R. A. Friesner, and R. M. Levy, Comparative Performance of Several


Flexible Docking Programs and Scoring Functions: Enrichment Studies for a Diverse Set of
Pharmaceutically Relevant Targets, J. Chem. Inf. Model., vol. 47, no. 4, pp. 15991608,
2007.

[9] AutoDock AutoDock. [Online]. Available: http://autodock.scripps.edu/. [Accessed: 15-


Jul-2014].

[10] Home of David S. Goodsell. [Online]. Available: http://mgl.scripps.edu/people/goodsell/.


[Accessed: 15-Sep-2014].

[11] M. Lape, C. Elam, and S. Paula, Comparison of current docking tools for the simulation of
inhibitor binding by the transmembrane domain of the sarco/endoplasmic reticulum calcium
ATPase, Biophys. Chem., vol. 150, no. 13, pp. 8897, Aug. 2010.

[12] MGLTools Website - Welcome MGLTools. [Online]. Available:


http://mgltools.scripps.edu/. [Accessed: 15-Sep-2014].

[13] M. Olk, J. Filipovi, and M. Prokop, FastGridThe Accelerated AutoGrid Potential Maps
Generation for Molecular Docking, Comput. Inform., vol. 29, no. 6+, pp. 13251336, 2012.

[14] Laboratory for Molecular Design and Drug Discovery. [Online]. Available:
http://www.rosalindfranklin.edu/cop/Home/LaboratoryforMolecularDesignandDrugDiscovery.
aspx. [Accessed: 25-Sep-2014].

[15] C. Ma, R. Kotaria, J. A. Mayor, S. Remani, D. E. Walters, and R. S. Kaplan, The Yeast
Mitochondrial Citrate Transport Protein CHARACTERIZATION OF TRANSMEMBRANE
DOMAIN III RESIDUE INVOLVEMENT IN SUBSTRATE TRANSLOCATION, J. Biol.
Chem., vol. 280, no. 3, pp. 23312340, Jan. 2005.

[16] C. A. Nicolaou and N. Brown, Multi-objective optimization methods in drug design, Drug
Discov. Today Technol., vol. 10, no. 3, pp. e427e435, Sep. 2013.

[17] P. H. Palestro, L. Gavernet, G. L. Estiu, and L. E. Bruno Blanch, Docking Applied to the
Prediction of the Affinity of Compounds to P-Glycoprotein, BioMed Res. Int., vol. 2014, p.
e358425, May 2014.

[18] W. Duch, K. Swaminathan, and J. Meller, Artificial intelligence approaches for rational drug
design and discovery, Curr. Pharm. Des., vol. 13, no. 14, pp. 14971508, 2007.

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 49
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

[19] E. Yu, Agent-oriented modelling: software versus the world, in Agent-Oriented Software
Engineering II, Springer, 2002, pp. 206225.

[20] G. D. Crnkovic, The Cybersemiotics and Info-Computationalist Research Programmes as


Platforms for Knowledge Production in Organisms and Machines, Entropy, vol. 12, no. 4,
pp. 878901, Apr. 2010.

[21] B. Chandrasekaran, J. R. Josephson, and V. R. Benjamins, What are ontologies, and why
do we need them?, IEEE Intell. Syst., vol. 14, no. 1, pp. 2026, 1999.

[22] A. H. Andr Filipe de Moraes Batista, Emerson Aguiar Noronha, Fbio Arago da Silva,
Maria das Graas Bruno Marietto, Robson dos Santos Frana Terry Lima Ruas, Modeling
Artificial Life Through Multi-Agent Based Simulation. INTECH Open Access Publisher, 2011.

[23] M. University of Southampton Luck and AgentLink, Agent technology: computing as


interaction; a roadmap for agent based computing. Southampton: University of
Southampton on behalf of AgentLink III, 2005.

[24] Agent Technology: Computing as Interaction,A Roadmap for Agent Based


Computing.[Online]. Available: http://www.das.ufsc.br/~jomi/mas/leituras/agentlink-
roadmap.pdf [Accessed: 16-Sep-2014]. .

[25] D. J. Cook, Multi-agent smart environments, J. Ambient Intell. Smart Environ., vol. 1, no. 1,
pp. 5155, 2009.

[26] Q. Guo and M. Zhang, Research on intelligent manufacturing system based on multi-
agent, in Intelligent Robotics and Applications, Springer, 2008, pp. 829838.

[27] Z. Li, C. H. Sim, and M. Y. H. Low, A survey of emergent behavior and its impacts in agent-
based systems, in Industrial Informatics, 2006 IEEE International Conference on, 2006, pp.
12951300.

[28] Wooldridge M. and Jennings N. Intelligent Agents: Theory and Practice.1995. .

[29] G. Weiss, Multiagent systems a modern approach to distributed artificial intelligence.


Cambridge, Mass.: MIT Press, 1999.

[30] An Ontology for Engineering Mathematics. [Online]. Available: http://www-


ksl.stanford.edu/knowledge-sharing/papers/engmath.html. [Accessed: 03-Aug-2015].

[31] M. N. Huhns and M. P. Singh, Ontologies for agents, Internet Comput. IEEE, vol. 1, no. 6,
pp. 8183, 1997.

[32] F. Zini and L. Sterling, (1) DISI-Universit a di Genova via Dodecaneso 35, 16146 Genova,
Italy zini@ disi. unige. it (2) Department of Computer Science & Software Engineering.

[33] P. Maret, J. Calmet, and others, Agent-based knowledge communities, Int. J. Comput. Sci.
Appl., vol. 6, no. 2, pp. 118, 2009.

[34] Molecular_Docking_Tutorial.pdf. .

[35] Machado_et_al_LNCS_4643_1-11_2007.pdf. .

[36] Carboxypeptidase. [Online]. Available:


http://www.chemistry.wustl.edu/~edudev/LabTutorials/Carboxypeptidase/carboxypeptidase.h
tml. [Accessed: 17-Sep-2014].

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 50
M.Harindra R. Fernando, Asoka S. Karunananda & Roshan P. Perera

[37] Potential Energy Function. [Online]. Available:


http://www.ch.embnet.org/MD_tutorial/pages/MD.Part2.html. [Accessed: 17-Sep-2014].

[38] Electric energy and potential. [Online]. Available:


http://physics.bu.edu/~duffy/PY106/Potential.html. [Accessed: 25-Oct-2014].

[39] P. Leeson, Drug discovery: Chemical beauty contest, Nature, vol. 481, no. 7382, pp. 455
456, Jan. 2012.

[40] J. W. Ponder and D. A. Case, Force fields for protein simulations, Adv. Protein Chem., vol.
66, pp. 2785, 2003.

[41] Jade Site | Java Agent DEvelopment Framework. [Online]. Available: http://jade.tilab.com/.
[Accessed: 17-Sep-2014].

[42] FIPA ACL Message Structure Specification. [Online]. Available:


http://www.fipa.org/specs/fipa00061/SC00061G.html. [Accessed: 18-Jul-2014].

[43] Apache Jena - Home. [Online]. Available: https://jena.apache.org/. [Accessed: 09-Sep-


2014].

[44] RDF Current Status - W3C. [Online]. Available:


http://www.w3.org/standards/techs/rdf#w3c_all. [Accessed: 29-Jul-2014].

[45] Y. Yan, C. Wang, A. Zhou, W. Qian, L. Ma, and Y. Pan, Efficiently querying rdf data in triple
stores, in Proceedings of the 17th international conference on World Wide Web, 2008, pp.
10531054.

[46] J. C. Arprez, A. Gmez-Prez, A. Lozano-Tello, and H. S. A. N. Pinto, Reference ontology


and (ONTO) 2 agent: the ontology yellow pages, Knowl. Inf. Syst., vol. 2, no. 4, pp. 387
412, 2000.

[47] SPARQL Query Language for RDF. [Online]. Available: http://www.w3.org/TR/rdf-sparql-


query/. [Accessed: 25-Jul-2014].

[48] Jmol: an open-source Java viewer for chemical structures in 3D. [Online]. Available:
http://jmol.sourceforge.net/. [Accessed: 17-Sep-2014].

[49] D. Rudenko and A. Borisov, An overview of blackboard architecture application for real
tasks, in Scientific Proceedings Of Riga Technical University, Ser, 2007, vol. 5, pp. 5057.

[50] T. J. Ewing and I. D. Kuntz, Critical evaluation of search algorithms for automated
molecular docking and database screening, J. Comput. Chem., vol. 18, no. 9, pp. 1175
1189, 1997.

[51] Locating Binding Sites in Protein Structures. [Online]. Available:


http://www.chemcomp.com/journal/sitefind.htm. [Accessed: 08-Sep-2014].

[52] L. Jirkovsk, S. M. Mak, and I. Kolingerov, Finding Cavities in a Molecule.

[53] Latest pairings | IUPHAR/BPS Guide to PHARMACOLOGY. [Online]. Available:


http://www.guidetopharmacology.org/latestPairings.jsp. [Accessed: 17-Sep-2014].

International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (7) : Issue (3) : 2016 51